Admissions 2030: The Intelligent Front Door.

Automation handles a defined, repeatable process with clear rules. AI comes in when the data isn't structured and needs judgement applied to it. Most of the confusion in the sector comes from people hearing "AI" and picturing ChatGPT, when it's a much broader toolkit than that.Lesley O'Keeffe

Most institutions are aiming for the top of the maturity map: full, joined-up, intelligent automation. In reality, based on Niico's own work and JISC/EY survey data, most institutions are still at the earliest stages. There's a long way to go before that top stage is realistic for most.Alistair Sergeant

The sector wants to change, but often doesn't know how. Much of the resistance comes down to language: a lack of understanding of what AI or agents actually do, versus what they're assumed to do. When people hear "AI", they think ChatGPT, when it's a much broader toolkit. There's a genuine piece of work to be done in upskilling and education before mindsets shift, and that has to happen alongside giving teams the headspace to engage with it, not on top of an already relentless workload.Lesley O'Keeffe

The honest answer: it may reduce headcount at some institutions, and pretending otherwise given current mergers and cuts isn't credible. But the argument isn't "AI replaces people". It's freeing people from repetitive work so they can focus on the admissions decisions that genuinely need a human: the borderline cases, the difficult conversations, the ones that need real care.Ben Rogers & Lesley O'Keeffe

Yes. Lesley's own institution has a mitigating circumstances agent live today, checking a genuinely multi-system process. In the first few weeks it handled around 1,000 applications the team never had to touch manually. The advice: start small and tangible, build trust in the result, then scale, rather than trying to automate an entire process in one move.Lesley O'Keeffe

Three principles. Baseline before you start, two or three weeks of volume, response time and backlog data is enough and it costs nothing, and most ROI cases collapse because nobody measured the before. Measure the work, not the people, headcount is the wrong unit and the one everyone's afraid of. Our own reference point at ARU: 700 enquiry emails a day reduced to roughly 50 needing a human, a capacity number, not a redundancy number. Say the intent out loud on day one, whether that's holding headcount through rising volume or genuinely reducing roles, ambiguity is what people fill with fear, and staff will work it out either way. One thing that buys real credibility: publish the failure modes alongside the wins, escalation rate, what the system got wrong, what humans had to fix, teams trust the numbers far more when you show the misses too. And be honest about the real comparison: for most institutions it isn't AI versus the current team, it's AI versus the same team absorbing 20% more volume with no budget for more staff.Alistair Sergeant

Money is tight almost everywhere, and most institutions carry a legacy of bruising, badly-run change projects, so waiting for your existing provider is understandable. The drawback is speed: providers build for the average case, not for what makes your institution different, so real innovation gets lost in the wait. Starting a procurement process today likely means a 2030 go-live at best, by which point the technology has moved on again. The alternative the paper argues for: incremental automation layered onto what you already run, rather than one large-scale replacement project.Lesley O'Keeffe & Ben Rogers

Yes. Niico works with institutions from a few hundred students up to several thousand, and the technology is built to scale to that size. The starting point is the same regardless of institution size: find one process worth automating that stacks up on time saved and ROI. Being smaller does not put this out of reach.Alistair Sergeant

It depends on where admissions sits at your institution, but realistically it needs all three working in partnership. IT support is non-negotiable for any successful implementation, but the drive has to come from the business side too, or projects stall on anxiety rather than momentum. Executive-level backing matters for smoothing over the inevitable bumps along the way.Ben Rogers

That accountability doesn't go away, and it shouldn't. The processes worth automating first are the well-defined, high-volume ones where a human stays firmly in charge of anything genuinely borderline or difficult. The agent does the heavy lifting; the accountable person still makes the call that matters.Ben Rogers & Lesley O'Keeffe

The goal isn't to avoid dependency, every institution is already dependent on its SRS, its CRM, its finance system. The goal is to make that dependency reversible. Own three things regardless of supplier: your data, exportable on demand in open formats, without a professional-services charge to get it out; your process logic, documented outside the vendor's configuration screen so it could be rebuilt elsewhere; and your team's capability, if nobody can still do the work manually, you haven't automated a process, you've forgotten one. Keep AI as a layer sitting on top of your systems of record rather than becoming the record itself, if the layer goes, the record still stands. Ask any supplier two questions: are you locked to a single model provider, or can you swap it without a full re-implementation, and what are the exit terms and the cap on annual price escalation? Then apply the degradation test: if this disappeared the week before Clearing, is the honest answer "slower and more expensive", or "we stop"? If it's the second, you've over-integrated. Start with work where failure is survivable, enquiry handling and document checks, not offer decisions.Alistair Sergeant

They're not necessarily in competition. For generic information, it may not matter to an applicant whether they're speaking to a bot or a person. The more interesting opportunity sits at the deep-personalisation end: AI that spots when a piece of research or news matches an applicant's stated interests, and surfaces that automatically, may actually go further than peer-to-peer communication can on its own.Lesley O'Keeffe

Yes, explicitly. The paper isn't arguing for removing face-to-face conversations for the decisions that need real judgement. What changes is everything around them: the chasing, the document checks, the repeat questions, freeing up time and headspace for the conversations that actually need a person.Ben Rogers

Quality rarely degrades from volume itself, it degrades for two separate reasons, and they need separate answers. Mismatch: growth harms quality when you recruit students who were never going to succeed on that course. The fix is measurement discipline, track continuation and completion by recruitment channel and entry route, not just applications and offers, a channel that converts brilliantly and continues badly is costing you money and outcomes at the same time. Capacity strain: growth plans should be stress-tested against staff-to-student ratios, timetable and estate, placement availability, assessment load and student support demand, not just admissions throughput, and where the honest limit is reached, cap the intake. In England, the B3 thresholds on continuation, completion and progression make this concrete, growth that breaks those isn't growth. Where AI helps: selection fit, consistency, speed of decision making, and early identification of at-risk students in the first term. Where it doesn't help: manufacturing teaching capacity. Worth also considering modes that don't load the same infrastructure, modular and LLE-funded provision, online delivery, degree apprenticeships, rather than piling more bodies into the same lecture theatres.Alistair Sergeant